融合KL散度和移地距离的高斯混合模型相似性度量方法

Similarity measure method of Gaussian mixture model by integrating Kullback-Leibler divergence and earth mover's distance

为提高高斯混合模型(GMM)间相似性度量方法的计算效率和准确性,通过对称化KL散度(KLD)并结合移地距离(EMD)提出一种新的相似性度量方法。首先计算待比较的两个高斯混合模型内各高斯成分间的KL散度,对称化处理后用于构造地面距离矩阵;然后用线性规划方法求解两个高斯混合模型间的移地距离作为高斯混合模型间的相似性度量。实验结果表明,将该相似性度量方法应用于彩色图像检索,相对于传统方法能够提高检索的时间效率和准确性。

To improve the computation efficiency and effectiveness of the similarity measure method between two Gaussian Mixture Models （GMM）, a new measure method was proposed by means of integrating symmetrized Kullback-Leibler Divergence （KLD） and earth mover＇s distance. At first, the KL divergence between Gaussian components of the two GMMs to be compared was computed and symmetrized for constructing the earth distance matrix. Then, the earth mover＇s distance between the two GMMs was computed using linear programming and it was used for GMM similarity measure. The new measure method was tested in colorful image retrieval. The experimental results show that the proposed method is more effective and efficient than the traditional measure methods.